Search code examples
numpydeep-learningobject-detectionmobilenettensorflow-lite

Mobilenet SSD input shape

I have trained a mobile SSD V2 model on a custom dataset for a single object detection task, and I have converted it to tflite. When I load the .tflite model using the interpreter for testing, and geting the input details using : input_details = model.get_input_details(), it outputs 

[{'name': 'normalized_input_image_tensor',
'index': 272,
'shape': array([  1, 300, 300,   3], dtype=int32),
'dtype': numpy.uint8,
'quantization': (0.0078125, 128)}]

I understand that the "300x300" are the images height and width, "3" is for the RGB channel, but to what the first element ("1") is referring ?

Solution

  • Shape: [Batch_size, height, width, channel]

    If you want to change that size, you need to set up it before converting into the pb file, for example using export_inference_graph